{"name":"Vihang Patel","url":"https://vihang.com/","updated":"2026-10-13","attribution":"Self-published by Vihang Patel. Source records carry their own verification qualification; a source check is not independent corroboration of every claim.","profile":{"name":"Vihang Patel","location":"Singapore","role":"Vihang Patel is based in Singapore and leads the Data & AI business unit at Marketnode as Vice President, Head of Data & AI, since 1 November 2025.","themes":[{"id":"value","title":"Where enterprise AI value actually lands","description":"Economics, adoption and exception work."},{"id":"trust","title":"Trust in automated work","description":"Abstention, evidence and review by exception."},{"id":"adoption","title":"How regulated institutions adopt AI","description":"Governance, partners and distribution."},{"id":"leadership","title":"Leading through technology shifts","description":"Organisation design and developing leaders."}],"leadership":["At Finaxar, I led a lending and payments business across four markets, as CEO until January 2021 and then Chairman. The organisation reached a peak of 100+ people, with executive functional heads and country leaders responsible for their own teams.","I have held P&L responsibility at company and business-unit level. Today I lead Marketnode’s Data & AI business unit, including its P&L, product and sales, reporting to the CEO. The work centres on enterprise AI adoption in regulated financial services.","At Splore, I led heads of technology, product, AI/ML and MLOps, and sales through the company’s acquisition by Model ML. Earlier, at Crayon Data, I led a 35+ person engineering organisation working on an enterprise data platform."],"careerOutcomes":["Menyala: Resident Builder at Temasek’s venture studio. I took Splore from project kickoff to greenlight in about four months, aligning senior studio and investment stakeholders. Splore became Menyala’s first exit; customers, product, IP and part of the team moved to Model ML.","Splore: I led functional heads for technology, product, AI/ML and MLOps, and sales. The technology group had 10 engineers; the product lead managed two product managers and a designer, and the AI group had two engineers. Recurring revenue per client grew fourfold.","Finaxar: I led a lending and payments business operating across four markets, as CEO until January 2021 and then Chairman. The organisation reached a peak of 100+ people across five countries, including the India development centre. Functional heads and country leaders ran their own teams. Monk’s Hill Ventures and 500 Startups were publicly reported backers; Cathay Financial Holdings as an investor is owner-reported. The Vietnam partnership involved Cathay Financial, Indovina Bank and Finaxar; my description of the bank relationship is Cathay United Bank with Indovina Bank.","Marketnode: I report to the CEO and lead product, engineering and business work in Data & AI. Smartflow cleared a Tier-1 bank security and risk review. These are owner-reported outcomes; measurement details and independent corroboration are not published here."],"facts":["Vihang Patel is based in Singapore and leads the Data & AI business unit at Marketnode as Vice President, Head of Data & AI, since 1 November 2025.","Previously: Splore, CEO & Director, February 2023–2025 (project kickoff November 2022); Menyala, Resident Builder, July 2022–January 2023.","Finaxar: Co-Founder, May 2016–July 2022; CEO to January 2021, then Chairman.","Dragon Wealth Asia: CTO & Head of Product, May 2014–April 2016; Operating Advisor at Monk’s Hill Ventures, October 2014–October 2022.","Avagam: Founder & CEO, July 2010–March 2012; Crayon Data: Technology Leader, April 2012–April 2014.","Earlier: Deloitte, risk management and banking consulting, 2005–2009; Numind, strategy consulting, 2009–2010.","Education: Master of Science and Bachelor of Science (Honours), Mathematics & Computing, IIT Kharagpur. INSEAD Global Management Programme (CGM), alumnus; Harvard Business School High Potentials Leadership Program (executive education)."],"smartflowOutcome":"95%+ of fields straight through, the rest routed to human review. Owner-reported; independent confirmation pending.","qualification":"First-party career account by Vihang Patel. Company and team achievements retain their attribution. Linked catalog records carry their own source and verification qualifications; a source check is not independent corroboration of every claim.","links":{"about":"https://vihang.com/about/","work":"https://vihang.com/work/","writing":"https://vihang.com/writing/","projects":"https://vihang.com/projects/","elsewhere":"https://vihang.com/elsewhere/","ask":"https://vihang.com/ask/","record":"https://vihang.com/record.json"}},"essays":[{"slug":"enterprise-ai-economics","title":"Where does enterprise AI value go?","url":"https://vihang.com/writing/enterprise-ai-economics/","label":"Essay","description":"A successful automated case can be cheap. The business case still depends on the share of cases using the system, how often exceptions occur, and how much work those exceptions create.","theme":"Where enterprise AI value actually lands","published":"2026-10-12","updated":"2026-10-12","kind":"Interactive essay · synthetic scenarios","scopeNote":"The scenarios are synthetic and exclude implementation costs; released capacity needs an operating decision before it becomes financial value.","notesSince":"2026-03","interactiveModel":"https://vihang.com/projects/ai-economics/","sections":[{"heading":"The cheap answer is only part of the cost","paragraphs":["I tend to ask who will do the work an AI system leaves behind. A wrong field, an ambiguous document or an unresolved exception still needs somebody’s time. A small inference bill can sit beside a large review queue.","Brynjolfsson, Li and Raymond’s 2023 NBER study examined 5,179 customer-support agents and found uneven productivity gains across workers. It is a useful warning against carrying one uplift across an organisation. The process and the people using the tool matter."]},{"heading":"The same model can make or lose money","paragraphs":["Consider 1,000 cases a month, 20 manual minutes per case and labour at $60 an hour. The manual baseline is $20,000. With 60% adoption, 25% of adopted cases needing 30 minutes of review, $1 AI usage per adopted case and $2,000 fixed monthly cost, the model gives $4,900 net benefit and 125 hours of released capacity.","Hold the volume and costs steady. At 85% adoption, 10% review and 20 review minutes, the benefit rises to $12,450. At 75% adoption, 60% review and 50 review minutes, it becomes a $10,250 loss and adds 125 hours of work. The difference comes from the work around the answer."]},{"heading":"Fewer exceptions can still cost more","paragraphs":["At $60 an hour, a 25% chance of a 30-minute review costs $7.50 per adopted case. In the starting scenario, that is more than the $1 AI usage cost. The useful question is how much attention each exception consumes.","Five percent needing two hours costs $6 per adopted case; twenty percent needing fifteen minutes costs $3. Automating the easy work can leave a smaller queue of harder cases. I would measure the spread of resolution times as well as the average.","More adoption scales the economics already present. If the value per adopted case is positive, adoption helps spread fixed costs. If it is negative, adoption grows the loss. A review bottleneck can also lengthen turnaround even when the average cost looks attractive."]},{"heading":"Capacity needs somewhere useful to go","paragraphs":["The 125 hours in the starting case are an operating choice. They could support more throughput, shorter turnaround or less overtime. Turning them into cash savings needs a staffing or demand decision.","For a stable, repetitive process with little review, a simple cost-per-case comparison may be enough. Before carrying that result elsewhere, I would check eligible-work adoption, intervention frequency and the time needed to resolve an exception. Those three observations tell me more than the price of a successful demonstration."]},{"heading":"How the calculation works","paragraphs":["Manual hours = monthly cases × manual minutes ÷ 60. Review hours = adopted cases × review rate × review minutes ÷ 60. Net benefit subtracts remaining manual labour, review labour, AI usage and fixed cost from the manual baseline.","Break-even adoption is fixed cost divided by monthly cases and the value per adopted case. With zero fixed cost and negative per-case value, only zero adoption breaks even; with zero fixed cost and zero per-case value, every adoption level breaks even. “Not reachable” means the model finds no break-even level between zero and 100%.","The calculation assumes constant volume, average labour rates and linear costs. Adopted cases without exceptions need no further handling in this model. Routine human handling across all adopted cases needs a separate cost estimate; review minutes apply only to cases counted as exceptions. Implementation, transition costs, undetected quality failures and changing demand need a separate estimate."]}],"sources":[{"title":"Brynjolfsson, Li and Raymond: Generative AI at Work, NBER working paper","url":"https://www.nber.org/papers/w31161","checked":"2026-10-12","published":"2023-04-24"}],"revisions":[{"version":"launch","date":"2026-10-12","label":"Launch version","summary":"Original publication, before the editorial revision.","url":"https://vihang.com/writing/enterprise-ai-economics/versions/launch/"},{"version":"2026-10-12-editorial","date":"2026-10-12","label":"Editorial revision","summary":"Clearer claims and counterarguments, before the incident-based revision.","url":"https://vihang.com/writing/enterprise-ai-economics/versions/2026-10-12-editorial/"},{"version":"2026-10-12-notes","date":"2026-10-12","label":"Prior version","summary":"Published text before the current rewrite.","url":"https://vihang.com/writing/enterprise-ai-economics/versions/2026-10-12-notes/"},{"version":"current","date":"2026-10-12","label":"Current version","summary":"Added observed incidents, explicit evidence and technical tradeoffs; original publication date unchanged.","url":"https://vihang.com/writing/enterprise-ai-economics/"}],"attribution":"Original self-published piece by Vihang Patel"},{"slug":"completion-needs-readback","title":"Test the part that does the work","url":"https://vihang.com/writing/completion-needs-readback/","label":"Essay","description":"A fluent reply can hide a broken tool path. I learned to test an agent at the boundary its work depends on.","theme":"Trust in automated work","published":"2026-10-12","updated":"2026-10-12","kind":"Engineering essay","notesSince":"2026-06","evidence":"https://vihang.com/agent-office-evidence.json","visual":{"title":"Three different tests","alt":"A reply reaches the model. A tool-using test crosses the model and command runner. An independent outcome check compares the resulting artifact with the intended state. A missing helper can leave the reply check green while stopping the work.","caption":"Each check answers a new question. The most useful small test crosses the boundary the work depends on.","image":"https://vihang.com/images/verification-boundaries.svg"},"sections":[{"heading":"The reply was fine. The command runner was missing.","paragraphs":["A coding agent gave a plausible reply in a smoke test. The same package failed when it had to read files or run a command: its command-host helper was missing. The first test had exercised the model while leaving the working part of the service untouched.","The repair included the helper in the package and checked it when the agent launched. On 1 October 2026, the follow-up test ran a repository command and compared the answer with the actual repository. That was a small test, but it crossed the boundary I depended on."]},{"heading":"Choose the question before choosing the test","paragraphs":["For a summariser, the quality of the text may be the result. For an agent that changes a record, the question is whether the right record changed. For one that starts a job, the question is whether the job finished. A single success flag cannot answer all three.","I find it useful to separate the reply, the tool action and the resulting work. This keeps a good explanation from covering a failed action. It also makes the failure easier to locate: was the input wrong, did the tool fail, or did another system show an older value?"]},{"heading":"A retry can create more work","paragraphs":["A timeout leaves an awkward question: did the operation fail, or did the response fail to reach me? Sending the request again may create a duplicate. Waiting may leave the original work unfinished.","AWS describes this problem in its guidance on idempotent APIs. A caller supplies an identifier for the intended request so a retry can be recognised as the same operation. That identifier has to preserve intent; reusing it for a different action creates a different error.","A delayed view needs another response. Amazon EC2 documents cases where a newly created resource temporarily appears absent. Its guidance checks the resource state with increasing waits between attempts. Retrying the creation and waiting for the existing resource are different choices."]},{"heading":"The extra check has a cost","paragraphs":["Checking every layer adds time, API calls and more failure paths. For a short task, a committed record returned by the tool may be enough. For a high-cost change, I want to inspect the result where it will be used. Repeated low-cost work may justify sampling once the process is understood.","The choice depends on the cost of a missed error and how quickly it would be found. A check that merely repeats the model’s explanation buys little. A check of an independent record can answer a new question."]},{"heading":"Make the test do useful work","paragraphs":["I now start with the smallest task that uses the service’s essential capability: run a command, retrieve a cited document or change a test record and inspect it. Then I add the failure I am concerned about, such as a missing dependency, a delayed response or a repeated request.","The practical lesson is to test the working path early. A convincing answer is cheap evidence when the job depends on something else."]}],"sources":[{"title":"Anthropic: Demystifying evals for AI agents","url":"https://www.anthropic.com/engineering/demystifying-evals-for-ai-agents","checked":"2026-10-12","published":"2026-01-09"},{"title":"AWS: Eventual consistency in the EC2 API","url":"https://docs.aws.amazon.com/ec2/latest/devguide/eventual-consistency.html","checked":"2026-10-12"},{"title":"AWS Builders’ Library: Making retries safe with idempotent APIs","url":"https://aws.amazon.com/builders-library/making-retries-safe-with-idempotent-APIs/","checked":"2026-10-12"}],"revisions":[{"version":"launch","date":"2026-10-12","label":"Launch version","summary":"Original publication, before the editorial revision.","url":"https://vihang.com/writing/completion-needs-readback/versions/launch/"},{"version":"2026-10-12-editorial","date":"2026-10-12","label":"Editorial revision","summary":"Clearer claims and counterarguments, before the incident-based revision.","url":"https://vihang.com/writing/completion-needs-readback/versions/2026-10-12-editorial/"},{"version":"2026-10-12-notes","date":"2026-10-12","label":"Prior version","summary":"Published text before the current rewrite.","url":"https://vihang.com/writing/completion-needs-readback/versions/2026-10-12-notes/"},{"version":"current","date":"2026-10-12","label":"Current version","summary":"Added observed incidents, explicit evidence and technical tradeoffs; original publication date unchanged.","url":"https://vihang.com/writing/completion-needs-readback/"}],"attribution":"Original self-published piece by Vihang Patel"},{"slug":"where-authority-belongs","title":"Who owns the work when agents hand it over?","url":"https://vihang.com/writing/where-authority-belongs/","label":"Essay","description":"Agents can divide a task quickly. Keeping one person responsible for the result takes more care.","theme":"Trust in automated work","published":"2026-10-12","updated":"2026-10-12","kind":"Architecture essay","notesSince":"2026-07","visual":{"title":"Keep an owner through the handoff","alt":"Analyst, executor and verifier contribute a proposal, action and independent evidence to one shared case, which holds the owner and next move. Repair and human decisions keep an owner; a verified result completes the case.","caption":"Specialists can change while the obligation and its next owner remain on one record.","image":"https://vihang.com/images/owned-work.svg"},"sections":[{"heading":"A handoff should carry an obligation","paragraphs":["In my Agent Office, research, execution and review can happen in different sessions. Each agent can produce a useful contribution while the overall task waits. The work record is what tells me who needs to move next.","This is familiar from running teams. Passing a document to a colleague is easy; making the next decision and its owner clear is the harder part. An agent handoff needs the same care, because its conversation may end long before the obligation does."]},{"heading":"Separate access from responsibility","paragraphs":["Multica holds the commitment and its current owner. The MCP portal supplies access to shared tools. Model routing chooses the reasoning capability. These are three different decisions, and changing one should leave the others clear.","The June 2025 MCP authorization specification describes access to protected servers through OAuth. That answers an access question. I still need a work record that says what was requested, who can carry it forward and when the owner should be involved.","A specialist may need broad context to understand a problem and a narrow set of actions to carry out the decision. Giving every agent the same tools is convenient, but makes it harder to tell which handoff changed the responsibility."]},{"heading":"Keep the reviewer outside the action","paragraphs":["For work that can change important state, I separate the analyst, executor and verifier. The analyst prepares the evidence. The executor proposes and applies an authorised action. The verifier inspects the result independently.","The separation costs time. It also gives the reviewer something concrete to inspect: the proposed change, its supporting evidence and the version that will be used. A second agent repeating the first agent’s reasoning would add conversation without adding much protection."]},{"heading":"A refusal needs a way forward","paragraphs":["A review can stop an incorrect action and still leave the task stranded. I want the reason and the next owner on the same work record. The owner may need to correct an input, choose another route or decide to stop.","This is where the operating model earns its cost. It keeps recovery attached to the original commitment, rather than creating a second conversation that everyone assumes somebody else is handling."]},{"heading":"Use the smallest organisation that works","paragraphs":["One agent is often enough for a short task with a clear owner and an easily checked result. I use more structure when work spans days, depends on several specialists or changes something expensive to repair.","The test is whether coordination reduces the attention I spend recovering lost work. A larger roster is a poor substitute for a clear next move."]}],"sources":[{"title":"MCP: HTTP authorization specification","url":"https://modelcontextprotocol.io/specification/2025-06-18/basic/authorization","checked":"2026-10-12","version":"2025-06-18"},{"title":"RFC 8693: OAuth 2.0 Token Exchange","url":"https://www.rfc-editor.org/rfc/rfc8693","checked":"2026-10-12","version":"February 2020"},{"title":"SLSA: Verifying artifacts","url":"https://slsa.dev/spec/v1.2/verifying-artifacts","checked":"2026-10-12","version":"1.2"}],"revisions":[{"version":"launch","date":"2026-10-12","label":"Launch version","summary":"Original publication, before the editorial revision.","url":"https://vihang.com/writing/where-authority-belongs/versions/launch/"},{"version":"2026-10-12-editorial","date":"2026-10-12","label":"Editorial revision","summary":"Clearer claims and counterarguments, before the incident-based revision.","url":"https://vihang.com/writing/where-authority-belongs/versions/2026-10-12-editorial/"},{"version":"2026-10-12-notes","date":"2026-10-12","label":"Prior version","summary":"Published text before the current rewrite.","url":"https://vihang.com/writing/where-authority-belongs/versions/2026-10-12-notes/"},{"version":"current","date":"2026-10-12","label":"Current version","summary":"Added observed incidents, explicit evidence and technical tradeoffs; original publication date unchanged.","url":"https://vihang.com/writing/where-authority-belongs/"}],"attribution":"Original self-published piece by Vihang Patel"}],"work":[{"slug":"smartflow","title":"Smartflow","url":"https://vihang.com/work/smartflow/","angle":"AI in regulated banking operations","company":"Marketnode","summary":"A document field becomes useful when its evidence, review and operational meaning stay connected.","role":"Vihang Patel is based in Singapore and leads the Data & AI business unit at Marketnode as Vice President, Head of Data & AI, since 1 November 2025.","chapter":{"id":"document-ai","label":"V","variable":"Trust","principle":"Abstain and route; judge critical fields separately."},"problem":"Credit-agreement onboarding connects complex documents to operational systems.","technical":["Smartflow's public material describes field-level lineage and human-in-the-loop review. The technical question is how a reviewer can inspect both a proposed value and its supporting clause, especially when amendments change the relevant terms. A well-formed field is not, by itself, evidence that it came from the controlling document.","Abstention is a useful design principle when evidence is missing or conflicting: route the unresolved value for a decision rather than fill it with a plausible guess.","Evaluate accepted-field precision, coverage, intervention effort and the final reviewed result separately. A final corrected result and the share of work completed automatically answer different operating questions. Confidence calibration is another distinct question: Guo and colleagues’ research provides a method for checking whether confidence estimates agree with observed correctness."],"technicalSources":[{"title":"Smartflow public product description","url":"https://ai.marketnode.com/","checked":"2026-10-12"},{"title":"Guo et al.: On Calibration of Modern Neural Networks","url":"https://proceedings.mlr.press/v70/guo17a.html","published":"2017","checked":"2026-10-12"}],"judgment":"My responsibility covers the operating model around the product: what can pass through, what needs human review, how a bank adopts the system and how the business supports delivery. Those decisions connect product quality to commercial and operating responsibility.","delivery":"I lead the Data & AI business unit’s P&L, product and sales, reporting to the CEO. The published Finastra partnership and Loan IQ demo provide the public account of the integration.","outcomes":["The published partnership connects Smartflow with Finastra’s Loan IQ. Finastra describes the potential to reduce credit-agreement onboarding time."],"related":{"title":"The economics of enterprise AI","url":"https://vihang.com/writing/enterprise-ai-economics/"},"sources":[{"id":"2026-finastra-marketnode-release","type":"press","entity":"Marketnode","title":"Finastra and Marketnode transform credit agreement onboarding through AI","issuer":"Finastra / Marketnode / PR Newswire","date":"2026-04-08","url":"https://www.finastra.com/press-media/finastra-and-marketnode-transform-credit-agreement-onboarding-through-ai","shows":"Strategic partnership integrating Smartflow with Loan IQ; \"can reduce\" credit-agreement onboarding time","qualification":"source-verified 2026-10-10"},{"id":"2026-loaniq-marketnode-demo","type":"product","entity":"Marketnode","title":"Loan IQ & Marketnode demo video: Automated loan onboarding","issuer":"Finastra","date":"2026-02-27","url":"https://www.finastra.com/viewpoints/videos/demo/loan-iq-marketnode-demo-video-automated-loan-onboarding","shows":"Public demo of the Smartflow and Loan IQ integration","qualification":"source-verified 2026-10-10"},{"id":"smartflow-site","type":"product","entity":"Marketnode","title":"Smartflow: every credit event extracted, structured, governed","issuer":"Marketnode","date":null,"url":"https://ai.marketnode.com/","shows":"Product site with published outcomes","qualification":"source-verified 2026-10-10"},{"id":"2026-loan-ops-room","type":"byline","entity":"Marketnode","title":"The Loan Ops Room: Clean Data Is the Starting Line","issuer":"Marketnode","date":"2026-08","url":"https://ai.marketnode.com/resources/","shows":"Article by Vihang","qualification":"source-verified 2026-10-10"}],"attribution":"First-party account by Vihang Patel; company and team results keep their attribution."},{"slug":"splore","title":"Splore","url":"https://vihang.com/work/splore/","angle":"From semantic search to source-cited research","company":"Menyala → Splore","summary":"Retrieval became an institutional product when the answer carried evidence a reader could inspect.","role":"Previously: Splore, CEO & Director, February 2023–2025 (project kickoff November 2022); Menyala, Resident Builder, July 2022–January 2023.","chapter":{"id":"retrieval","label":"IV","variable":"Understanding","principle":"Every answer points to its source."},"storyHeading":"Semantic search before ChatGPT","story":["From July to October 2022 at Menyala, we built and tested a live semantic-search MVP with a small contracted team. The starting point was transformer embeddings and accessible consumer information, including gaming data. Real users judged retrieval relevance on their own queries. This preceded Splore’s formal kickoff in November 2022.","The early product gave us a way to test relevance before taking on harder unstructured material. The work moved towards enterprise financial services and narrowed to private markets, where useful information was often outside a public search index. It was the access problem I had encountered at Dragon Wealth, now at institutional scale.","In 2023–2024, we built research agents on a custom harness for source-cited portfolio research and post-deal compliance. The engineering question expanded from finding relevant material to checking whether the answer stayed faithful to it. In 2024, embedding work included fine-tuning open models from the Gemma 2 and BGE families.","The product served institutional asset managers and a private-equity firm. Splore obtained ISO 27001 certification and SOC 2 attestation. I led functional heads for technology, product, AI/ML and MLOps, and sales as the product and commercial work developed.","I led Splore through its acquisition by Model ML. Customers, product, IP and part of the team carried forward. The durable principle for me is that a reader should be able to follow an answer back to the material supporting it."],"outcomes":["Splore became Menyala’s first exit. The linked speaking appearance records my public work during that period."],"related":{"title":"Work","url":"https://vihang.com/work/"},"sources":[{"id":"2024-sff-fireside","type":"talk","entity":"Splore","title":"Future of Fund Management with Technology (SFF 2024 fireside)","issuer":"Singapore FinTech Festival","date":"2024-11","url":"https://luma.com/hlg9h9vt","shows":"Fireside chat speaker","qualification":"Owner-reported; independent confirmation pending."}],"attribution":"First-party account by Vihang Patel; company and team results keep their attribution."},{"slug":"finaxar","title":"Finaxar","url":"https://vihang.com/work/finaxar/","angle":"Regulated lending across four markets","company":"Finaxar","summary":"Merchant data could support instant credit decisions. Larger lines brought the work back to documents and review.","role":"Finaxar: Co-Founder, May 2016–July 2022; CEO to January 2021, then Chairman.","chapter":{"id":"credit","label":"III","variable":"Trust and access","principle":"Autonomy is earned per decision, and depends on the stakes."},"storyHeading":"Move the decision boundary with better evidence","story":["Finaxar’s cloud lending platform used merchant data from partners including Lazada and Stripe to make instant credit-line decisions. Larger lending decisions needed evidence beyond the data available through platform APIs. Bank statements and financial documents brought extraction and human judgment back into the process.","That changed how I thought about automation. The amount at risk and the evidence available belonged in the same decision. A fast decision on a small line did not settle how to assess a larger exposure.","As co-founder and CEO, I held operating and P&L responsibility. Functional heads and country leaders ran technology, product, risk, sales, finance and people. Finaxar operated in Singapore, Hong Kong, Vietnam and Malaysia: four markets. The organisation reached a peak of 100+ people across five countries, including the India development centre. I was CEO until January 2021, then Chairman until July 2022."],"outcomes":["Public announcements document the leadership team, the Visa collaboration, the Lazada partnership and the Vietnam bank partnership."],"related":{"title":"Company recognition","url":"https://vihang.com/elsewhere/#recognition"},"sources":[{"id":"2020-flex-visa","type":"press","entity":"Finaxar","title":"Flex by Finaxar, powered by Visa","issuer":"Visa / PYMNTS","date":"2020-10-15","url":"https://www.visa.com.sg/about-visa/newsroom/press-releases/flex-by-finaxar-powered-by-visa-helps-smes-in-southeast-asia-control-and-reconcile-expenditures-via-a-unified-spend-management-solution.html","shows":"Spend-management product jointly developed with Visa","qualification":"source-verified 2026-10-10"},{"id":"2020-finaxar-leadership","type":"press","entity":"Finaxar","title":"Singapore-based Finaxar strengthens leadership team with three key appointments","issuer":"Jumpstart / Hubbis","date":"2020-05-04","url":"https://www.jumpstartmag.com/singapore-based-finaxar-strengthens-leadership-team-with-three-key-appointments/","shows":"Senior hires into the executive team with banking and investment leadership experience","qualification":"source-verified 2026-10-11"},{"id":"2019-finaxar-cathay-indovina","type":"press","entity":"Finaxar","title":"Cathay Financial, Indovina Bank, Finaxar tie-up to improve SME financing in Vietnam","issuer":"The Business Times / DealStreetAsia","date":"2019-07-17","url":"https://www.businesstimes.com.sg/international/asean/cathay-financial-indovina-bank-finaxar-tie-improve-sme-financing-vietnam","shows":"Three-way partnership for SME financing in Vietnam","qualification":"source-verified 2026-10-11"},{"id":"2018-lazada-finaxar","type":"press","entity":"Finaxar","title":"Lazada works with Finaxar to offer financing for online vendors","issuer":"KrASIA","date":"2018-08-24","url":"https://kr-asia.com/alibabas-lazada-works-with-finaxar-to-offer-financing-for-online-vendors","shows":"Alibaba/Lazada financing partnership","qualification":"source-verified 2026-10-10"},{"id":"2019-visa-everywhere","type":"award","entity":"Finaxar","title":"Visa Everywhere Initiative 2019, Singapore winner","issuer":"Visa","date":"2019-03-21","url":"https://www.visa.com.sg/about-visa/newsroom/press-releases/visa-names-finaxar-as-winner-of-visa-everywhere-initiative-2019-in-singapore.html","shows":"Winner, Singapore finals","qualification":"source-verified 2026-10-10"},{"id":"2019-mas-fintech-awards","type":"award","entity":"Finaxar","title":"MAS FinTech Awards 2019, Singapore Founder, 2nd (FXR One)","issuer":"MAS","date":"2019-11-11","url":"https://www.mas.gov.sg/-/media/mas/news/media-releases/2019/annex-b---about-fintech-awards.pdf","shows":"Runner-up; lending-as-a-service platform for banks","qualification":"source-verified 2026-10-10"}],"attribution":"First-party account by Vihang Patel; company and team results keep their attribution."},{"slug":"crayon-data","title":"Crayon Data","url":"https://vihang.com/work/crayon-data/","angle":"Enterprise data and engineering leadership","company":"Avagam → Crayon Data","summary":"The data preparation, inference and retrieval around a billion-node graph mattered as much as the model.","role":"Avagam: Founder & CEO, July 2010–March 2012; Crayon Data: Technology Leader, April 2012–April 2014.","chapter":{"id":"data-and-wealth","label":"II","variable":"Understanding and access","principle":"Ground truth comes first."},"storyHeading":"Build the data system around the model","story":["I founded Avagam and continued the work after its acquisition into Crayon Data. As Technology Leader, reporting to the CEO, I led a 35+ person engineering organisation working on graph and machine-learning capabilities.","The billion-node graph made the surrounding engineering visible. Data processing, inference and retrieval determined whether a model could become a usable product. The model was one part of the system we had to build and operate.","The platform offered multi-tenant SaaS and dedicated single-tenant cloud deployments. Those options addressed different enterprise data requirements while keeping the delivery model in the cloud. At Dragon Wealth, the same question became more specific: how to turn financial documents into facts a recommendation could use."],"outcomes":["Crayon Data was a CODE_n top-50 finalist at CeBIT in March 2014. The platform was later recognised as a TiE50 winner in May 2014, after I left in April."],"related":{"title":"Dragon Wealth: from fund documents to portfolios","url":"https://vihang.com/work/dragon-wealth/"},"sources":[{"id":"2014-cebit-coden","type":"selection","entity":"Crayon Data","title":"CODE_n top-50 finalist, CeBIT 2014","issuer":"Times of India / CIOL","date":"2014-03","url":"https://timesofindia.indiatimes.com/tech-news/crayon-data-in-final-list-for-big-data-award/articleshow/31868747.cms","shows":"Finalist and the only Asian company; not the winner","qualification":"source-verified 2026-10-10"},{"id":"2014-tie50","type":"award","entity":"Crayon Data","title":"TiE50 winner, TiECon 2014","issuer":"TiE / Business Today","date":"2014-05","url":"http://blogs.intoday.in/businesstoday/Crayon-Data-startup-gets-global-recognition-63138.html","shows":"Recognised the platform he led architecture and core IP for (awarded after he moved on in April 2014)","qualification":"source-verified 2026-10-10"}],"attribution":"First-party account by Vihang Patel; company and team results keep their attribution."},{"slug":"dragon-wealth","title":"Dragon Wealth","url":"https://vihang.com/work/dragon-wealth/","angle":"From fund documents to portfolios","company":"Dragon Wealth Asia","summary":"Fund documents and structured data became personalised recommendations and model portfolios on web and mobile.","role":"Dragon Wealth Asia: CTO & Head of Product, May 2014–April 2016.","chapter":{"id":"data-and-wealth","label":"II","variable":"Understanding and access","principle":"Ground truth comes first."},"storyHeading":"Make the recommendation traceable to its inputs","story":["At Dragon Wealth, I led product and technology for a cloud wealth platform on AWS. It combined fund documents with complex structured data to support personalised recommendations and model portfolios, delivered through web and mobile applications.","The recommendation depended on the facts underneath it. Fund information had to be available, extracted and organised before it could become an input to portfolio construction. Retail-relevant information was difficult to obtain from public sources, so access and understanding were part of the product problem.","This experience connects the data-platform work at Crayon Data with the retrieval work that later became Splore. A useful interface can make a recommendation look simple; the work is in making its inputs dependable enough to use."],"outcomes":["Euromoney named Dragon Wealth Best Asian non-bank for innovation in its 2014 Innovations in Wealth Management Technology awards."],"related":{"title":"Splore: source-cited research","url":"https://vihang.com/work/splore/"},"sources":[{"id":"2014-euromoney","type":"award","entity":"Dragon Wealth","title":"Innovations in Wealth Management Technology: Best Asian non-bank for innovation","issuer":"Euromoney","date":"2014-11-04","url":"https://www.euromoney.com/article/27bjsstsqxhkmh0wsdjox/awards/innovations-in-wealth-management-technology-best-asian-non-bank-for-innovation/","shows":"Awarded to Dragon Wealth during his tenure as CTO & Head of Product","qualification":"source-verified 2026-10-10"}],"attribution":"First-party account by Vihang Patel; company and team results keep their attribution."},{"slug":"bank-consulting","title":"Bank consulting","url":"https://vihang.com/work/bank-consulting/","angle":"Risk and technology programmes","company":"Deloitte · Numind","summary":"Bank risk and strategy work made the evidence behind a decision as important as the decision itself.","role":"Earlier: Deloitte, risk management and banking consulting, 2005–2009; Numind, strategy consulting, 2009–2010.","chapter":{"id":"bank-risk","label":"I","variable":"Trust","principle":"Institutions trust what they can audit."},"storyHeading":"Start with the evidence an institution can audit","story":["At Deloitte, I worked on banking risk management; at Numind, on banking and financial-technology strategy. The work crossed Europe, Asia and the Middle East, with different teams responsible for risk, technology and delivery.","Those programmes gave me an early view of the trust problem. A model or proposed process had to fit the evidence an institution could inspect. The people taking responsibility for a decision needed a way to question its basis.","I return to that problem in document AI and agent systems. A result needs supporting evidence, and the person responsible for it needs a practical route for reviewing exceptions. Better automation changes the mechanism; it leaves that responsibility in place."],"outcomes":["Deloitte from 2005–2009 and Numind from 2009–2010."],"related":{"title":"About","url":"https://vihang.com/about/"},"sources":[],"attribution":"First-party account by Vihang Patel; company and team results keep their attribution."}],"projects":[{"slug":"agent-office-build-log","title":"Agent Office build log","url":"https://vihang.com/projects/agent-office-build-log/","kind":"Build log","summary":"Selected design decisions, repairs and releases from May to October 2026, beginning with the shared harness precursor.","attribution":"Independent project by Vihang Patel on personal time and infrastructure."},{"slug":"this-site-is-an-mcp-server","title":"This site is an MCP server","url":"https://vihang.com/projects/this-site-is-an-mcp-server/","kind":"Public interface","summary":"Published pages and their sources through read-only MCP and WebMCP tools; one content set serves readers and assistants.","attribution":"Independent project by Vihang Patel on personal time and infrastructure."},{"slug":"ai-economics","title":"AI economics explorer","url":"https://vihang.com/projects/ai-economics/","kind":"Interactive tool · synthetic","summary":"Change adoption, exception work and costs to see what happens to capacity and the operating model. Inputs stay in the page.","attribution":"Independent project by Vihang Patel on personal time and infrastructure."},{"slug":"governed-agent-control-plane","title":"Agent Office","url":"https://vihang.com/projects/governed-agent-control-plane/","kind":"Operating experiment","summary":"How work keeps an owner across agent handoffs, and why operating failures changed the tests I trust.","attribution":"Independent project by Vihang Patel on personal time and infrastructure."},{"slug":"how-ai-assistants-describe-me","title":"How AI assistants describe me","url":"https://vihang.com/projects/how-ai-assistants-describe-me/","kind":"Evaluation protocol","summary":"A protocol for checking identity, currency, source support and attribution across assistant surfaces.","attribution":"Independent project by Vihang Patel on personal time and infrastructure."}]}